Skill Injection Detection on SkillHarm FPP setting
55.6Detection RateSkill Scanner
Evaluation Results
| Method | Links | |
|---|---|---|
| Skill ScannerBackbone=Opus 4.72026.06 | 55.6 | |
| Skill ScannerBackbone=GPT-5.52026.06 | 21.5 | |
| Agent ScanBackbone=Own API2026.06 | 7.4 |
Loading the SOTA2 catalog…